We have all seen Spot, the autonomous four-legged robot created by US robotics firm Boston Dynamics, run, jump, and dance to famous tunes. Now it turns out that the robot dog can also make for a pretty good tour guide.
In a video posted by the Hyundai-owned company, a team of engineers is shown to have outfitted the robot with a top hat, mustache, googly eyes, and fake eyebrows as it talks with them in a British accent, also taking them on a tour of the company’s Massachusetts facility.
LLMs Give Spot the Ability to Make Conversations and Recognize Objects and Places
The tour starts with Spot walking itself to the charging station where other Spot robots go to rest and recharge their batteries. The robot dog calls the area the “first point of interest”.
To make Spot more conversational, Boston Dynamics fitted its system with APIs from OpenAI’s ChatGPT and other open-source large language models (LLMs) before carefully training its responses. Then the engineers attached a speaker to its body, added text-to-speech capabilities, and made it possible for Spot’s mouth to mimic speech like the mouth of a puppet.
According to Matt Klingensmith, the principal software engineer at Boston Dynamics, the team gave Spot a very brief script detailing each room at the facility that it would visit. Then with help from its generative AI assistants, the bot combines data from the script with images captured by built-in cameras on its head and body to get more information about what it was seeing and generate responses accordingly.
Boston Dynamics noted that Spot uses a machine-learning algorithm called the Visual Question-Answering model to essentially answer questions based on subjects captured by its cameras.
ALSO READ:- Blockchain Technology’s Future Unfolds: Join WOW Summit Dubai 2023
‘Spot’ was Able to Take Multiple Personas and Answer Questions Sarcastically
The most interesting part of the video is when Spot assumes multiple personas throughout the video. At the start, the robot dog is a fancy British butler and a few minutes in takes on the personality of a 1920s archaeologist, then a teenager, and finally an actor from one of Shakespeare’s plays. It even took the form of a sarcastic guide, which, when asked to come up with a haiku, said “Generator hums low in a room devoid of joy. Much like my soul.”
The Boston Dynamics team was also impressed by its ability to generate responses to random questions. Midway through the tour, the team asks Spot who its parents are, and the robot takes them over to an area of the facility where older Spot models are displayed. In another instance, its LLM cracked jokes by suggesting that Stretch, another robot developed by the company that is designed to move heavy objects, was made to do yoga.
ALSO READ:- Google Wallet to Add ORCA Card Support, Simplifying Transit Payments
AI Enables Robots to Understand Tasks in a Real World Context, Also Helps Reduce Learning Curve
In a blog post on the Boston Dynamics’ website, Klingensmith wrote that the company was excited to explore the “intersection” of artificial intelligence and robotics as LLMs can help provide “cultural context, general commonsense knowledge, and flexibility” that could be helpful while performing several robotics tasks. He highlighted a possible scenario where users will be able to assign tasks to Spot just by talking, thereby reducing the learning curve for using these systems.
Klingensmith noted that AI would make it possible for robots to not just follow orders but in “some sense” understand the actions they take in the context of the world around them. He also added that generative AI systems could be useful for applications that haven’t been realized yet.
Speaking on whether robots like Spot could put human tour guides out of work, the lead engineer of Boston Dynamics said that AI chatbots still have the tendency to make stuff up, a phenomenon called “hallucinations” in the machine-learning world, which would make it prone to falsifying historical facts and figures. The functioning of generative AI technologies also depends on the strength of the internet connection, added Klingensmith.
